GOOGLE WOKE UP LATE TO THE AI CODING WAR

Añadido: 2026-05-27

[00:00:00]So, Google just dropped a brand new Anti-Graffiti 2, and this is their take to try to get into AI coding tools war that is majorly being fought with Cloud, Cursor, and Codex.

[00:00:15]Because right now, when people usually think about AI coding agents, they usually think about Win Surf, Cursor, Codex, Copilot even, but they do not usually think about Google. And this is Google trying to get into this marketplace.

[00:00:32]The weirdest part is why Google was so far back into the AI coding tools war.

[00:00:39]Because they got the users, they got the models, they got the infrastructure that needs to run these agents. So, why it was so far back? Which Google actions make they did not take the lead into this race?

[00:00:53]So, what is Anti-Graffiti 2? Google is positioning it as a agentic platform to make AI coding.

[00:01:00]And it's basically what most of the companies are delivering nowadays. A chatbot that has agentic AI that make two calls into the system and control it with bash.

[00:01:17]We have here the projects management and the sessions that these projects have, just like Codex or Cloud desktop is delivering right now.

[00:01:28]The biggest catch they got right now is the the Gemini 3.5 Flash model. It's the brand new development model.

[00:01:40]They are selling these as one of the most capable models they have right now, even surpassing 3.1 Pro in a bunch of benchmarks.

[00:01:52]And Google is describing Anti-Graffiti 2 as a new developer desktop application for semi customizing and orchestrating code agents. And you can see here, we can do a lot of development work here and simultaneously talk with multiple agents. And it's more like an orchestrator than anything else.

[00:02:17]The biggest thing about the new into gravity is the Gemini new model flash that is not flash anymore. Google keeps calling these model flash, but it definitely has its own categoric now, especially when we think about the pricing change that is the major upgrade they have done here in this model.

[00:02:40]Back in the Gemini 2.0 flash era, the pricing was basically very aggressive compared to the models that we had in the market around 10 cents per million input and 40 cents per million output. It was very aggressive and they keep bumping up to each launch they had in the last few in the flash category.

[00:03:08]Now surpassing the $1.5 per million input and $9 output. It's almost getting to the GPT 5.4 price category. If we get here in the GPT 5.3 codex, we can see that the input price is almost the same. The output here is definitely way higher than the Google, but to compare the from the first flash model that was the two, it is very more close to the GPT 5.3 codex and the GPT 5.4 pricing category than it is to the first flash model that Google has ever launched.

[00:03:56]So, what is the thing about this model?

[00:03:58]It's definitely not as intelligent as is the Gemini 3.1 pro model because they this definitely a way more bigger model in parameters, but it's definitely way better in the agentic benchmarks, and we can see here through the terminal bench that it scored significantly higher than the Gemini 3.1 Pro, and especially the flash version. So, the real focus of these new model is the agentic workflows, and the agentic development.

[00:04:36]Here in the artificial analysis, we can get uh pretty much deep on into the pricing range, and especially into the intelligence range. And I will say this launch of 3.5 flash is especially very good talking about the price range. We can see here it has basically their own category into the sort of models.

[00:05:01]It's not close to 4.7, and not even closer to 5.5, but it's not the claim. They are trying to make these their own category definitely, but the thing is is one of the fastest models that we have right now into this sort of category because it's way better than GPT 5.5, even in the low version, and especially into the flash models like uh GPT-4 flash is it has the most token per minute I put tokens per second I ever seen in a model close to the Kimik K2 6, and the Grok 4.3 and the Gemini 3.5 is almost close to the Nvidia Nemotron that was one of the fastest models we had that has launched it lately.

[00:05:59]So, yes, even though it's way uh expen- more expensive compared to the flash models, it has the benefit, especially talking about intelligence and coding capabilities.

[00:06:12]Uh the price it's up so to the tokens per second that we are having with this model. We can see here that is uh multi uh capability model.

[00:06:24]And we can see here in the open house that the inputs that are they receive is text, image, video, file, audio.

[00:06:35]And output is usually text, but yes, it's a multi uh capability model that has a range of inputs. So, yes, it has the share of the market and it's definitely one of the best sort of models that we have right now.

[00:06:57]Google has also announced the anti-gravity CLI and said in the shares is the same agents and as as anti-gravity 2.

[00:07:05]Uh two version. The Google developer blogs also say gravity to gravity CLI is available to everyone, while some customers Gemini CLI and Gemini code assist extension path to stops seven requests on June 18.

[00:07:21]So, I I I believe this is one of the biggest launch, especially because the Gemini CLI was terrible. It was not usable at all. I did not like the Gemini CLI, but this one is specially good.

[00:07:37]It's clean. It's fast compared to the Gemini CLI, it's definitely an improvement.

[00:07:44]So, the biggest improvement in this new model, I would say, is definitely the TPS. It's a It's an incredibly fast model. It is very capable, especially into the speed we can see here. The amount of the cool two cows they do each moment is absurd compared to GPT-5.5 or Cloud Code that's are very slow models.

[00:08:11]They Cuz they usually do a bunch of two cows at the same time, it looks fast. But, compared to the TPS, we can see the Gemini 3.5 is definitely one of the fastest.

[00:08:25]So, the question is not, "Isn't Gravity 2 better than Cursor today?" That's too small. What happens if Google finally to Gemini Cloud AI student Android workspace and developer tools into one series identical flow? We do not know yet. They are still working on it. But, because if they get this right, and Gravity stop being just another AI editor, it becomes Google attempted to own the future of the software development. And that was the launch matters. It will definitely say to us if the Google is still competitive and they are trying to take the lead into the AI coding war.

[00:09:06]And that's why this launch matters so much. Not because Google shipped another coding assistant. That's the boring take. The real story is that Google's trying to make Gemini the agent layer for developers. And if developers actually trust it, only the time and the market will show it. And this is the thing that gets interesting if it happens. Ademais is easy, hellability is really hard, and Google does not have it in the last releases. We have seen a lot of problems, especially talking about updates in the synchronizing version between all of the multiple coding uh AI coding tools. We had problems with the last launch that anti-gravity was not receiving the update with the new model.

[00:09:59]It was something very big and with these and with this new launch, they are trying to get it right because they are putting everything into the same harness that is the anti-gravity back back end that they have. They are putting the anti-gravity CLI too and the anti-gravity IDE using the same uh brain. So, probably this is the right step that they are taking into the AI coding war.

[00:10:30]But, Google does not win need to win the this demo. Google needs to win the workflow and this is the step that we will see into the next episodes into this AI war.

[00:10:45]So, thank you for watching and you can see more videos uh into our channel.

[00:10:49]Please subscribe and leave a like.

[00:10:52]Bye-bye.